In this paper, we consider the problem of maximizing the energy efficiency (EE) for multi-input multi-output (MIMO) interference channels, subject to the per-link power constraint. To avoid extensive information exchange among all links, the optimization problem is formulated as a noncooperative game, where each link maximizes its own EE. We show that this game always admits a Nash equilibrium (NE) and the sufficient condition for the uniqueness of the NE is derived for the case of arbitrary channel matrices, which can be checked in practice. To reach the NE of this game, we develop a totally distributed EE algorithm, in which each link updates its own transmit covariance matrix in a completely distributed and asynchronous way: Some players may update their solutions more frequently than others or even use the outdated interference information. The sufficient conditions that guarantee the global convergence of the proposed algorithm to the NE of the game have been given as well. We also study the impact of the circuit power consumption on the sum-EE performance of the proposed algorithm in the case when the links are separated sufficiently far away. Moreover, the tradeoff between the sum-EE and the sum-spectral efficiency (SE) is investigated with the proposed algorithm under two special cases: 1) low transmit power constraint regime; 2) high transmit power constraint regime. Finally, extensive simulations are conducted to evaluate the impact of various system parameters on the system performance.